The Glasgow Admission Prediction Score
Allan Cameron
Consultant Physician, Glasgow Royal Infirmary
Outline
The need for an admission prediction score
What is GAPS?
GAPS versus human judgment and Amb Score
GAPS as a predictor of adverse outcomes
The role of GAPS in ambulatory care
Mortality and admission odds against length of ED stay
Advantages of predicting admission
Identifying as early as possible which patients are
likely to be admitted and which are likely to be
discharged could promote efficiency:
Identifying patients for ambulatory care
Bed management
Decision support
Patient streaming
Triage is first clinical assessment made in ED
Triage staff cannot accurately predict admission
Background
Several tools have been created to predict admission
at the point of triage
The simpler tools lack accuracy
The accurate tools lack simplicity
We have lacked a simple but accurate tool to assess
the probability of admission at the time of triage
Glasgow admission prediction score
Variable Points
Age 1 point per decade
NEWS score 1 point per point on NEWS score
Triage category: 3 5
2 10
1 20
Referred by GP 10
Arrived in ambulance 5
Admission within 1 year 5
Methods
Multi-centre, retrospective, cross-sectional study
322,846 unscheduled secondary care attendances in
North Glasgow over a two-year period
Two-thirds of attendances were selected at random
to create the prediction score using variables already
available at triage
Score created from mixed-effects multiple logistic
regression model
The score was then tested for accuracy on the
remaining third by assessing its ROC curve
Results
344,429 adult attendances over 2 years
After discounting transfers between units and
missing data, 322,846 attendances were available for
analysis in 191,653 patients
123,397 of the 322,846 attendances led to admission
(38.22%)
215,231 attendances used to create the score
107,615 attendances used to test the score
Criticism
Do we really need another score? Whatever happened to
clinical judgement?
GAPS versus human judgment
Comparison of accuracy of triage nurses and GAPS
Prospective study of 1,838 ED attendances
Of these, 766 (41.7%) were admitted
Triage staff asked to estimate probability of admission (VAS)
Nurses were only accurate in predicting admission when they
were very confident of the outcome (92.4%) but accuracy was
poor in the majority of cases (68.8% accurate)
When the nurses were less confident, GAPS was significantly
more accurate and better calibrated
Criticism
This score predicts admissions. How can we use it to facilitate
ambulatory care? Don’t we already have a score for that?
GAPS versus Amb Score
Prospective study, GRI-led multi-site collaboration
Consecutive patients presenting for ED triage
Researchers worked in shifts to cover all 168 hours of
the week
Each patient interviewed to calculate GAPS and Amb
Scores
Patients followed up to 30 days
Endpoint was admission to hospital or ED discharge
Comparison of AUC of ROC using DeLong’s method
Results
1496 adults attending ED triage during study
Of these, 64 IRDs, leaving 1432 for analysis
570 (39.8%) admitted
AUC 0.808 for GAPS, compared to 0.743 for Ambs,
p<0.00001
GAPS had net classification improvement of 6% over
Amb
Criticism
Surely this just tells you whether someone will be admitted,
not whether they should be admitted?
We followed up all 1432 patients to six months
Split patients into equal 3 tertiles based on GAPS
Survival analysis using all-cause mortality
“Survival” analysis of hospital readmission
“Survival” analysis of hospital length of stay on the 563
patients admitted to hospital (7 lost to follow-up)
Prospective follow-up
p < 0.0001
p < 0.0001
p=0.026
Implementation
We have been using GAPS at our Acute Assessment Unit in
Glasgow Royal Infirmary for over two years
Of 1600 monthly GP referred medical attendances, around
30% can be sent directly to our ambulatory unit using the
single criterion of low GAPS (<25)
Achieves a high discharge rate from ambulatory first
assessment of ~90% with excellent safety record
Allows ambulatory care to be patient-based rather than
condition-based
GAPS has now been taken up by several UK sites
Conclusions
We have derived a simple but accurate way to assess
probability of admission at triage
It predicts death, reattendance and readmission
within 28 days
It usually outperforms experienced triage staff
It outperforms the current method recommended by
the RCP toolkit for streaming to ambulatory care
It can be used to measure (or control for) patient
factors when looking at admission rates
Further challenges
How can we better use the information GAPS gives
us in real time?
How can we use the information GAPS gives us for
service planning?
Dissemination and implementation.
www.FrontDoorMedicine.com
See the original papers and this slide set at